Study On Energy Trading Of Microgrid Based On Reinforcement Learning

The microgrid utilizes renewable energy to meet user's dynamic load requirements.Though energy trading can allocate surplus electricity,meet time-varying load demand,improve the utilization of renewable energy,reduce the dependence on non-renewable resources,such as coal,and improve benefits of microgrid.The paper studies the energy trading game among microgrids,and designs a microgrid energy trading system to deal with the limit power of microgrid.Thereinto,according to the time-varying renewable energy generation and power demand of the microgrids,microgrid negotiates the actual trading strategy with other connected microgrids,then perform the energy transaction.Besides,this paper deduces the Nash equilibrium and its existence condition of the energy trading game,and reveals the benefit of microgrid was influenced by the renewable energy production capacity,load demand and energy storage capacity.On this basis,this paper designs the energy trade algorithm based on Q-learning,micro-grid perceives renewable energy production capacity,load demand and current energy storage and other information with wireless network,no need to perceive the other connected microgrids' capacity and load model,the best trading strategy is achieved by a enough dynamic game.In addition,the paper presents a hotbooting Q-learning energy trading algorithm,a hotbooting process with emulational experiences is used to improve the exploration strength and convergence speed of the algorithm.Simulation shows that our scheme improves the utility of microgrid and reduces the dependence on non-renewable energy generation.For example,in a day trading 4 times,the hotbooting Q-learning trading algorithm increases utility of the microgrid by 5.31%and decreases the average unit of the energy purchased from the power plant by 33.33%compared with the Q-learning trading algorithm.In view of the numerous microgrid scenario,the paper designs a microgrid energy trading optimization scheme based on deep reinforcement learning.Through the observation information of renewable energy production capacity,load demand,current electricity quantity and the number of microgrid,quickly learn to grasp the change of external trading environment and obtain the optimal trading strategy.The algorithm can improve the utilization of renewable energy and benefits of microgrid.In order to boost learning and improve exploitation strength of DQN algorithm,a Fast DQN trading algorithm is proposed.Simulation shows that our scheme can obtain better energy trading utility of microgrid and achieve optimum power dispatching.For example,in a day trading 8 times,the DQN trading strategy increases the utility of microgrid by 53.57%compared with the hotbooting Q-learning strategy,and further increased by 4.65%with Fast DQN-based scheme.